A Load Forecasting Model for Power Systems Integrating Policy Analysis and Hybrid Neural Networks

Accurate load forecasting is essential for managing the new type of power system, particularly those integrating renewable energy sources. Traditional models often overlook the impact of policy changes, resulting in significant errors in systems with high renewable penetration. This paper presents a novel policy quantification system for power systems. The system employs Policy Modelling Consistency (PMC) and Term Frequency-Inverse Document Frequency (TF-IDF) to quantitatively assess policies. Additionally, it integrates the Analytic Hierarchy Process (AHP) method and Latent Dirichlet Allocation (LDA) models for qualitative analysis of policy texts. A new load forecasting model named PoliLoadNet is developed that features a hybrid architecture of residual network (ResNet), long-short-term memory (LSTM), and convolutional neural network (CNN). This model predicts load based on policy texts and other influential factors such as climate, industrial activity, and geographical location. Experimental results demonstrate that the model performs effectively in predicting loads in both long- and short-term loads, thereby enhancing the reliability and stability of the new type of power system.